Saraschandra KaranamHuman - AI Interactions Researcher · Enterprise AI
Back to Research
RESEARCH PROGRAM • 2007-11 & 2014-17

Cognitive Modeling of Web-Navigation and Information Search

Investigating the influence of cognitive factors affecting web-navigation/information search and modeling them.

Task DescriptionStartCurrent WebpagePrevious WebpageTarget Page?YesStopNoFocused-on AreaComprehension (LSA)Selection (Best Similarity)124AgeDomain KnowledgeNFCIncreasing similarity?YesIncrement path, clickNoBuffer: Maintains current pathCalculates path adequacy (PA)Stores old paths & adequaciesSpatial AbilityAgeIncreasing PA?YesNo33331BackRe-focusNext best
Representative Visualization

CoLiDeS+ Processing Model

Schematic diagram of steps involved in CoLiDeS+: Shaded circles indicate the locations where individual differences are involved (1: Age, 2: Domain Knowledge, 3: Spatial Ability and 4: Need for Cognition).

3Book Chapters
8Journal Articles
11Conference Papers
2Workshop Papers
5Universities
02 · WHY THIS RESEARCH MATTERED

Why this research mattered

Information search and web-navigation involve several cognitive processes such as memory, attention, problem solving, comprehension and decision making. These cognitive processes are known to be affected by one or more cognitive factors such as aging, domain knowledge and presence of graphical information on the pages. Understanding how the process of information search and navigation works and how it is influenced by one or more of the factors enables development of automated support tools that enhance performance.

RESEARCH QUESTION

"How to cognitively model the individual factors that influence web-navigation and information search?"

This research program investigated the cognitive processes underlying user navigation on websites and information search using search engines. This research developed and validated computational cognitive models to predict user clicks on websites, search query reformulations and clicks on search engine results.

Stream 01

Experiments, A/B Tests

Experiments investigating the influence of individual factors like age, domain knowledge and presence of graphical information on websites

Stream 02

Cognitive Modeling

Enhancing cognitive models with learnings from experiments to incorporate individual differences of age, domain knowledge and graphical information

Stream 03

Simulations

Run simulations of the enhanced models and compare with actual behavior

Stream 04

Build Support Tools

Build a predictive support tool using the enhanced computational cognitive models and evaluate their efficacy.

03 · EMPIRICAL DEVELOPMENT

Research Program

Connected studies conducted progressively answered different aspects of the central research question.

Study 01

CoLiDeS + Pic Model

Question

"How does graphical information on websites influence navigation behavior, and how to model it?"

Methods
A/B Tests - Behavioral ExperimentsCognitive ModelingSimulation
Outcome

Enhanced cognitive model of web-navigation that incorporates semantic information from pictures and a support tool for web-navigation based on the enhanced model

Study 02

Modeling individual differences (age, domain knowledge) in information search

Question

"How do age and domain knowledge influence information search behavior and how to model them?"

Methods
A/B Tests - Behavioral ExperimentsCognitive ModelingSimulation
Outcome

Enhanced cognitive model of information search that incorporates age and domain knowledge differences and a support tool for information search based on the enhanced model

04 · REPRESENTATIVE CASE STUDY

Role of Domain Knowledge in Cognitive Modeling of Information Search

Background

Cognitive models of information search (back then) did not incorporate individual differences in information search behaviour caused by differences in domain knowledge. This research filled that gap by first experimentally establishing that domain knowledge differences cause differences in information search behavior, enhanced the cognitive model to incorporate domain knowledge affects, ran simulations with the enhanced model and compared the simulations with actual user behavior.

Research Questions

  1. How to incorporate differences in domain knowledge levels of users into a computational cognitive model that predicts information search behavior?
  2. Would a model that takes differences in domain knowledge into account, predict user clicks on search engine result pages better than a model that does not?

Research Approach

Create two semantic spaces varying in the amount of domain knowledge (e.g., health)
Simulations on 6 information search tasks
Matching with actual behavioral data from 48 users (divided into high and low domain knowledge groups based on a domain knowledge test)
0.60.70.80.91.01.1Mean number of matches (per task)Non-ExpertExpertType of Semantic SpacePDKHighLow
Mean number of matches (per task) in relation to semantic space and prior domain knowledge (PDK)

Key Findings

01

Interaction of semantic space x domain knowledge of actual users

The efficacy of the modeling (in terms of the number of matches between model predictions and actual user clicks) was higher with the expert semantic space compared to the non-expert semantic space while for low domain knowledge participants it was the other way around

02

Interaction effect is lost if two different semantic spaces were not used

A plausible explanation for the interaction effect is that the expert and the non-expert semantic spaces give appropriate similarity values as assessed by users with high (more precise) and low (less precise) domain knowledge respectively

05 · PROGRAM SYNTHESIS

What We Learned

Learnings from this program

Effect of graphics/pictures

The accuracy of user navigation behavior is better when the semantic information from graphics/pictures on a web-page is highly relevant to the content of the page.

Effect of age

Older adults generate less search queries, use less keywords per query, reformulate less, spend longer time evaluating the search results, spend more time evaluating the websites opened and switch less often between search results and websites.

Effect of domain knowledge

Users with higher domain knowledge have more appropriate mental representations characterized by more relevant concepts, higher activation values, stronger connections between concepts. Users with higher domain knowledge therefore can comprehend the search results and content of websites better.

Enhanced cognitive models

Models that incorporate individual differences in search and navigation behavior due to differences in cognitive factors such as age and domain knowledge predict actual behavior with greater accuracy.

06 · RESEARCH ASSETS

Research Artifacts

Empirical assets and frameworks generated to guide future enterprise-wide design and engineering direction.

Cognitive models

CoLiDeS+Pic computational simulations mapping visual-semantic parsing. CoLiDeS+ - with variations that mimic individual differences in age and domain knowledge

LSA semantic spaces

Two semantic spaces with high and low health related information

Experimental results

Results of several A/B Tests investigating the influence of pictures, age and domain knowledge on web-navigation and information search

Support tools

Support tools based on the enhanced cognitive models

07 · PROGRAM PORTFOLIO

Program Portfolio

A portfolio of experiments, simulations and support tools exploring cognitive web-navigation and information search.

Experiment 01

Influence of text and graphics in locating web-page widgets

Experiment 02

Influence of picture icons next to hyperlinks

Experiment 03

Influence of text in main content in addition to hyperlinks

Experiment 04

Influence of text x relevance of pictures in addition to hyperlinks

Experiment 05

Influence of relevance of pictures on web-navigation

Cognitive Modeling 01

CoLiDeS + Pic: Enhanced cognitive model with semantic information from pictures

Simulation 01

Simulate and evaluate web-navigation behavior with enhanced model

Application 01

Using CoLiDeS + Pic for web-navigation support

Application 02

Using CoLiDeS + Pic for support in navigating 3D virtual environments

Experiment 06

Extend cognitive models to real websites

Experiment 07

Extend cognitive models to search engines

Experiment 08

Influence of domain knowledge on information search

Experiment 09

Influence of age on information search

Cognitive Modeling 02

Enhanced cognitive model with age-related differences

Application 03

Use enhanced cognitive model for support with information search